Model repeats itself in a loop

Symptom

The same phrase or sentence repeats until the token limit.

Likely causes

Temperature too low for the model
Context overflowed and truncation removed instructions
Wrong chat template or missing stop tokens

How to fix it

Raise temperature slightly (0.6 to 0.8) and add min-p 0.05.
Add a mild repeat penalty (1.05 to 1.1); strong values hurt coherence.
Check the prompt isn't truncated: increase the context size.
Verify the chat template and stop tokens.

bash

llama-cli -m model.gguf --temp 0.7 --min-p 0.05 --repeat-penalty 1.1 -c 8192

Still stuck? Check the lessons or measure first with the benchmark lab.